Lab2 数字图像频域增强实验¶
Author: Gao Pengbing (nbgao)
Email: nbgao@126.com
In [4]:
import numpy as np
import cv2
import matplotlib.pyplot as plt
import warnings
import math
from numpy.fft import fft2, fftshift, ifft2, ifftshift
%matplotlib inline
warnings.filterwarnings('ignore')
2D/3D 通用高斯低通滤波器¶
In [2]:
def dist(pa, pb):
dis = math.sqrt((pa[0]-pb[0])**2+(pa[1]-pb[1])**2)
return dis
def low_pass_filter(h, w, D):
mat_T = np.zeros((h,w))
center_point = (h/2, w/2)
for i in range(h):
for j in range(w):
dis = dist(center_point,(i,j))
mat_T[i,j] = np.exp(-(dis**2)/(2*(D**2)))
return mat_T
def GaussianLowPassFilter(image, D):
h, w = image.shape[:2]
if(np.ndim(image)==2): # 灰度图
fft = fft2(image)
fshift = fftshift(fft)
mat_T = low_pass_filter(h, w, D)
fshift_new = fshift*mat_T
img_fshift = np.log(np.abs(fshift_new))
img_new = np.uint8(np.abs(ifft2(ifftshift(fshift_new))))
return img_new, img_fshift
elif(np.ndim(image)==3): # 彩色图
fft_list = [fft2(image[:,:,i]) for i in range(3)]
fshift_list = [fftshift(fft_list[i]) for i in range(3)]
mat_T = low_pass_filter(h, w, D)
fshift_new_list = [fshift_list[i]*mat_T for i in range(3)]
img_fshift_list = np.log(np.abs(fshift_new_list))
img_new_list = [np.uint8(np.abs(ifft2(ifftshift(fshift_new_list[i])))) for i in range(3)]
img_new = cv2.merge(img_new_list[:])
img_fshift = cv2.merge(img_fshift_list[:])
return img_new, img_fshift
In [3]:
D_list = [5, 20, 50, 80, 250]
D_num = len(D_list)
def GaussianLowPassFilterResults(image):
if(np.ndim(image)==2): # 灰度图
plt.figure(figsize=(4,4))
plt.imshow(image, 'gray')
plt.title('Original Gray')
plt.axis('off')
plt.show()
plt.figure(figsize=(3*D_num, 2*3*image.shape[0]/image.shape[1]))
for i in range(len(D_list)):
img_new, img_fshift = GaussianLowPassFilter(image, D_list[i])
# 空域图像
plt.subplot(2, D_num, i+1)
plt.imshow(img_new, 'gray')
plt.title('D0={}'.format(D_list[i]))
plt.axis('off')
# 频域图像
plt.subplot(2, D_num, i+1+D_num)
plt.imshow(img_fshift, 'gray')
plt.axis('off')
plt.tight_layout()
plt.show()
elif(np.ndim(image)==3): # 彩色图
plt.figure(figsize=(4,4))
plt.imshow(image)
plt.title('Original RGB')
plt.axis('off')
plt.show()
plt.figure(figsize=(3*D_num, 4*3*image.shape[0]/image.shape[1]))
for i in range(len(D_list)):
img_new, img_fshift = GaussianLowPassFilter(image, D_list[i])
# 空域图像
plt.subplot(4, D_num, i+1)
plt.imshow(img_new)
plt.title('D0={}'.format(D_list[i]))
plt.axis('off')
# 频域图像 R通道
plt.subplot(4, D_num, D_num+i+1)
plt.imshow(img_fshift[:,:,0], cmap='Reds')
plt.axis('off')
# 频域图像 G通道
plt.subplot(4, D_num, 2*D_num+i+1)
plt.imshow(img_fshift[:,:,1], cmap='Greens')
plt.axis('off')
# 频域图像 B通道
plt.subplot(4, D_num, 3*D_num+i+1)
plt.imshow(img_fshift[:,:,2], cmap='Blues')
plt.axis('off')
plt.tight_layout()
plt.show()
In [4]:
for i in range(1,6):
file_path = '../Image/image'+str(i)+'.jpg'
img_bgr = cv2.imread(file_path)
img_rgb = cv2.cvtColor(img_bgr, cv2.COLOR_BGR2RGB)
img_gray = cv2.cvtColor(img_rgb, cv2.COLOR_RGB2GRAY)
print('Image{}:'.format(i), img_rgb.shape)
# 灰度图 空域图及频域图
# GaussianLowPassFilterResults(img_gray)
# 彩色图 空域图及频域图
GaussianLowPassFilterResults(img_rgb)
Work2-P2¶
自行选择一种频域的高通滤波器对彩色图像进行滤波操作, 取3组不同的参数进行实验，根据实验效果进行参数的比较分析
In [7]:
def dist(pa, pb):
dis = math.sqrt((pa[0]-pb[0])**2+(pa[1]-pb[1])**2)
return dis
def high_pass_filter(h, w, D):
mat_T = np.zeros((h,w))
center_point = (h/2, w/2)
for i in range(h):
for j in range(w):
dis = dist(center_point,(i,j))
mat_T[i,j] = 1 - np.exp(-(dis**2)/(2*(D**2)))
return mat_T
def GaussianHighPassFilter(image, D):
h, w = image.shape[:2]
if(np.ndim(image)==2): # 灰度图
fft = fft2(image)
fshift = fftshift(fft)
mat_T = high_pass_filter(h, w, D)
fshift_new = fshift*mat_T
img_fshift = np.log(np.abs(fshift_new))
img_new = np.uint8(np.abs(ifft2(ifftshift(fshift_new))))
# img_new = ifft2(ifftshift(fshift_new))
return img_new, img_fshift
elif(np.ndim(image)==3): # 彩色图
fft_list = [fft2(image[:,:,i]) for i in range(3)]
fshift_list = [fftshift(fft_list[i]) for i in range(3)]
mat_T = high_pass_filter(h, w, D)
fshift_new_list = [fshift_list[i]*mat_T for i in range(3)]
img_fshift_list = np.log(np.abs(fshift_new_list))
img_new_list = [np.uint8(np.abs(ifft2(ifftshift(fshift_new_list[i])))) for i in range(3)]
img_new = cv2.merge(img_new_list[:])
img_fshift = cv2.merge(img_fshift_list[:])
return img_new, img_fshift
In [20]:
D_list = [5, 10, 20]
D_num = len(D_list)
def GaussianHighPassFilterResults(image):
if(np.ndim(image)==2): # 灰度图
plt.figure(figsize=(4,4))
plt.imshow(image, 'gray')
plt.title('Original Gray')
plt.axis('off')
plt.show()
plt.figure(figsize=(3*D_num, 2*3*image.shape[0]/image.shape[1]))
for i in range(len(D_list)):
img_new, img_fshift = GaussianHighPassFilter(image, D_list[i])
# 空域图像
plt.subplot(2, D_num, i+1)
plt.imshow(img_new, 'gray')
plt.title('D0={}'.format(D_list[i]))
plt.axis('off')
# 频域图像
plt.subplot(2, D_num, i+1+D_num)
plt.imshow(img_fshift, 'gray')
plt.axis('off')
plt.tight_layout()
plt.show()
elif(np.ndim(image)==3): # 彩色图
plt.figure(figsize=(4,4))
plt.imshow(image)
plt.title('Original RGB')
plt.axis('off')
plt.show()
plt.figure(figsize=(4*D_num, 4*4*image.shape[0]/image.shape[1]))
for i in range(len(D_list)):
img_new, img_fshift = GaussianHighPassFilter(image, D_list[i])
# 空域图像
plt.subplot(4, D_num, i+1)
plt.imshow(img_new)
plt.title('D0={}'.format(D_list[i]))
plt.axis('off')
# 频域图像 R通道
plt.subplot(4, D_num, D_num+i+1)
plt.imshow(img_fshift[:,:,0], cmap='Reds')
plt.axis('off')
# 频域图像 G通道
plt.subplot(4, D_num, 2*D_num+i+1)
plt.imshow(img_fshift[:,:,1], cmap='Greens')
plt.axis('off')
# 频域图像 B通道
plt.subplot(4, D_num, 3*D_num+i+1)
plt.imshow(img_fshift[:,:,2], cmap='Blues')
plt.axis('off')
# plt.tight_layout()
plt.show()
In [21]:
for i in range(1,6):
file_path = '../Image/image'+str(i)+'.jpg'
img_bgr = cv2.imread(file_path)
img_rgb = cv2.cvtColor(img_bgr, cv2.COLOR_BGR2RGB)
img_gray = cv2.cvtColor(img_rgb, cv2.COLOR_RGB2GRAY)
print('Image{}:'.format(i), img_rgb.shape)
# 灰度图 空域图及频域图
# GaussianHighPassFilterResults(img_gray)
# 彩色图 空域图及频域图
GaussianHighPassFilterResults(img_rgb)
In [ ]: